類神經網路應用在土石流發生可能性分析

Abstract

近年來台灣集水區土石流災害頻傳，基此本研究嘗試透過內業處理的方式，取代部份土石流現場調查工作。期能提升土石流防治工作之效率，以為未來判定並提供集水區整治優選決策之參考。研究範圍鎖定南投縣一百九十九條潛勢溪流，應用地理資訊系統軟體─Arc Viewâ3.2為主軸，建立有關地質、水文及地文方面之土石流發生因子資料庫；再藉以類神經網路(ANN)判定土石流發生的可能性。分析結果顯示：訓練數據集正確率達89.2%；測試數據集正確率達83.4%。

Abstract In Taiwan, the frequency and magnitude of debris flow have both increased recent years. The government identified 1,420 watershed as high debris flow potential areas. However, the investigation and identification process is far from complete. Since most of the debris flow prone watersheds are in rural areas, the investigation of the natural conditions to assess the potential of occurrence has been difficult. It is crucial to establish a systematic approach using computer analytical tools to assess potential of debris flow more easily and more accurately. In this research a debris flow potential assessment model was developed. The geographic data of the watershed were analyzed by geographical information system (GIS) to enhance the accuracy and efficiency. A total of 199 stream watersheds in Nantou County were used as samples for developing and testing of the artificial neural network (ANN) assessment model. For each watershed, 13 parameters, representing the geographical, geological, hydrological condition of the watershed were analyzed by the artificial neural network to establish the model. The accuracy rate of the ANN model tested with the training data set and the simulation data set were 89.2% and 83.4%, respectively.